Source-linked AI summary
Zipf's law and log-normal distributions in measures of scientific output across fields and institutions: 40 years of Slovenia's research as an example
Matjaz Perc
TL;DR
Why citation patterns differ between publications and researchers, and why some researchers gain more attention, remains incompletely understood. Analyzing Slovenia’s research record, the paper finds power laws for publication citations but generally log-normal or sometimes exponential distributions for researcher indexes. The indexes have equivalent distributional properties, yet individual rankings can differ.
Problem
Why some researchers attract more attention than others remains unclear, and citation and researcher-output measures exhibit different distributional patterns.
Source: §1 · pp. 1–2Method
The paper analyzes Slovenia’s research output across institutions and fields, comparing publication-citation distributions with researchers’ g-index and h-index distributions and rankings.
Source: §2 · pp. 2–3, §3 · p. 3Results
Publication-citation power laws have α values from 2.41 to 3.13, while researcher-level g- and h-index distributions are generally log-normal or sometimes exponential, not power laws.
Source: §3 · pp. 4–6Takeaways & Limitations
The g- and h-index distributions are statistically equivalent, but different measures can rank individual researchers differently, supporting their use in assessing research quality and impact.
Source: §3 · pp. 6–7
Abstract
from arXiv · showhide
Slovenia's Current Research Information System (SICRIS) currently hosts 86,443 publications with citation data from 8,359 researchers working on the whole plethora of social and natural sciences from 1970 till present. Using these data, we show that the citation distributions derived from individual publications have Zipfian properties in that they can be fitted by a power law $P(x) \sim x^{-\alpha}$, with $\alpha$ between 2.4 and 3.1 depending on the institution and field of research. Distributions of indexes that quantify the success of researchers rather than individual publications, on the other hand, cannot be associated with a power law. We find that for Egghe's g-index and Hirsch's h-index the log-normal form $P(x) \sim \exp[-a\ln x -b(\ln x)^2]$ applies best, with $a$ and $b$ depending moderately on the underlying set of researchers. In special cases, particularly for institutions with a strongly hierarchical constitution and research fields with high self-citation rates, exponential distributions can be observed as well. Both indexes yield distributions with equivalent statistical properties, which is a strong indicator for their consistency and logical connectedness. At the same time, differences in the assessment of citation histories of individual researchers strengthen their importance for properly evaluating the quality and impact of scientific output.
1. Introduction
The paper motivates empirical study of citation patterns and researcher rankings as a way to understand scientific recognition and refine measures of research output. It contrasts power-law patterns in publication citations with log-normal or exponential patterns in researcher-level measures.
1. Introduction
- Researcher rankings matter for appointments and grants, while questions about why some researchers gain more recognition remain open.Scientific excellence alone does not explain all observed citation and output-index patterns.Source: §1 · p. 1
- Cumulative advantage and preferential attachment can produce power-law patterns, but deviations in citation data leave room for other explanations.The cited work also associates sublinear or near-linear preferential attachment with stretched-exponential or log-normal forms.Source: §1 · p. 2
- Across 40 years of Slovenian research, publication citations show power-law properties, whereas researcher-level g- and h-index distributions are log-normal or sometimes exponential.The indexes have equivalent distributional properties, yet can rank individual researchers differently.Source: §1 · p. 2
2. Preliminaries
The preliminaries establish SICRIS as the source for a documented national research record and describe the publication and researcher groups analyzed. They outline distributional fits and ranking comparisons used to examine citation patterns and output indexes.
2. Preliminaries
- SICRIS supports a well-documented research history for Slovenia, whose population is two million.Of 30,630 registered researchers, 8,359 have at least one Web of Science-indexed bibliographic unit.Source: §2 · p. 2
- The dataset contains 86,443 Web of Science-linked publications and 835,970 citations accumulated from 1970 onward.Bibliographies are updated continuously, and citation data are updated monthly via Web of Science.Source: §2 · p. 3
- Records were retrieved automatically and analyzed nationally and across selected institutions and fields, including Ljubljana, Jožef Stefan, medicine, and chemistry.The paper focuses on these representative examples while making larger tables available online.Source: §2 · p. 3
- The analysis compares citation and index distributions using Zipf plots, cumulative distributions, nonlinear fits, and maximum-likelihood and p-value checks.It also compares researchers’ relative ranks under the g-index, h-index, and total citation count.Source: §2 · p. 3
3. Results
Publication-citation distributions show power-law behavior across Slovenia, institutions, and fields, while researcher-level index distributions are generally log-normal or, in some settings, exponential. The g- and h-index distributions are statistically equivalent, although individual rankings can differ, especially lower in the rankings.
3. Results
- α ranges from 2.41 to 3.13 for publication-citation power laws across the analyzed sets.The Jožef Stefan Institute’s Zipf fit is less secure because it covers a considerably narrower region; the middle cumulative-distribution fit has p-value below 0.1.Source: §3 · p. 4, Figure 1 · p. 4
- Researcher-level g- and h-index distributions are not power laws; log-normal fits work generally, while exponential fits occur for Jožef Stefan and chemistry.The figure reports exponential decay parameters κ ≈ 14(1) and κ ≈ 8.7(3) for the middle and rightmost panels, respectively.Source: §3 · pp. 5–6, Figure 2 · p. 5
- Different output indicators yield differing researcher rankings, with deviations more likely among lower-ranked researchers than top-ranked researchers.The g- and h-index distributions are statistically equivalent, but individual citation records remain important to ranking within a group.Source: Figure 3 · p. 6, §3 · pp. 6–7
- Higher self-citation rates and Jožef Stefan’s historically hierarchical constitution are associated with the exponential researcher-index distributions.Chemistry has a reported self-citation rate of 0.31; the authors note that the reason for the observed distribution changes is difficult to pinpoint.Source: §3 · pp. 5–6
4. Summary
Citation counts per publication fit power laws, while researchers’ g- and h-index distributions are generally log-normal, with exponential forms in specified cases. The indexes have statistically equivalent distributions, but coauthor-blind national rankings can favor researchers in large collaboration networks.
4. Summary
- Publication citation distributions across Slovenia, institutions, and research fields fit power laws fairly accurately.The distributions include Slovenia as a whole.Source: §4 · p. 7
- Researchers’ g-index and h-index distributions are log-normal, or exponential for fields with high self-citation rates and organizations with special constitutions.Source: §4 · p. 7
- The g-index and h-index distributions are statistically equivalent, indicating consistency and logical connectedness between the indexes.Source: §4 · p. 7
- Ignoring coauthor counts in ARRS citation-based rankings creates favorable bias, with Particle Physics collaboration members dominating the rankings.Examples include DELPHI, Belle, and HERA-B.Source: §4 · p. 7
- The study may inform theoretical models of empirically observed distributions and encourage further attention to the topic.Source: §4 · p. 7